Skip to content

[BACKEND][LAYOUT] Use LL for AMDMfma related layout conversions #5210

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ThomasRaoux merged 10 commits into from
Nov 21, 2024
Merged

[BACKEND][LAYOUT] Use LL for AMDMfma related layout conversions #5210

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
e6a1758
Update
Jokeren Nov 21, 2024
ba11ef6
Update
Jokeren Nov 21, 2024
484a146
Update
Jokeren Nov 21, 2024
c5dbfb0
Merge branch 'main' into keren/mfma-ll
Jokeren Nov 21, 2024
63215c8
Update
Jokeren Nov 21, 2024
4cccf9a
Merge branch 'keren/mfma-ll' of github.com:openai/triton into keren/m…
Jokeren Nov 21, 2024
5d4ae07
Update
Jokeren Nov 21, 2024
b70faef
Update
Jokeren Nov 21, 2024
0286888
Update
Jokeren Nov 21, 2024
3eca786
Merge branch 'main' into keren/mfma-ll
ThomasRaoux Nov 21, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 11 additions & 11 deletions lib/Conversion/TritonGPUToLLVM/ConvertLayoutOpToLLVM.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -374,24 +374,24 @@ struct ConvertLayoutOpUsingLinearLayoutsConversion
// TODO (Keren): Currently, we handle general mma/blocked/slice/dot(ampere)
// -> mma/blocked/slice/dot(ampere) conversions. The following tasks must be
// completed before we can remove the layoutIsOK check:
// 1. Support for AMD's MFMA and WMMA
// 1. Support for AMD's WMMA
std::function<bool(Attribute)> layoutIsOK = [&](Attribute layout) {
if (auto nvidiaMma = dyn_cast<NvidiaMmaEncodingAttr>(layout)) {
if (useLegacyMMAConversion) {
return false;
}
return true;
if (isa<NvidiaMmaEncodingAttr, AMDMfmaEncodingAttr>(layout)) {
return !useLegacyMMAConversion;
}
if (auto dotOperand = dyn_cast<DotOperandEncodingAttr>(layout)) {
if (auto nvidiaMma =
dyn_cast<NvidiaMmaEncodingAttr>(dotOperand.getParent())) {
if (useLegacyMMAConversion) {
return false;
}
auto parent = dotOperand.getParent();
if (isa<MmaEncodingTrait>(parent) && useLegacyMMAConversion) {
return false;
}
if (auto nvidiaMma = dyn_cast<NvidiaMmaEncodingAttr>(parent)) {
if (nvidiaMma.isAmpere()) {
return true;
}
}
if (isa<AMDMfmaEncodingAttr>(parent)) {
return true;
}
return false;
}
if (isa<BlockedEncodingAttr>(layout)) {
Expand Down
2 changes: 2 additions & 0 deletions test/Conversion/amd/mfma-shortcut.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,7 @@ module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 4 :
tt.func public @shortcut_mfma16(%arg0: tensor<16x16xf16, #mfma>) {
// CHECK-NOT: store
// CHECK-NOT: load
// CHECK: llvm.return
%0 = triton_gpu.convert_layout %arg0 : tensor<16x16xf16, #mfma> -> tensor<16x16xf16, #dotop>
tt.return
}
Expand All @@ -21,6 +22,7 @@ module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 4 :
tt.func public @no_shortcut_mfma16(%arg0: tensor<16x16xf16, #mfma>) {
// CHECK: store
// CHECK: load
// CHECK: llvm.return
%0 = triton_gpu.convert_layout %arg0 : tensor<16x16xf16, #mfma> -> tensor<16x16xf16, #dotop>
tt.return
}
Expand Down